Tourism service method and system integrating local culture and tourism characteristics

Abstract

The invention belongs to the field of tourism planning, and provides a tourism service method and system integrating local culture and tourism characteristics, and the method comprises the steps: collecting local culture data; collecting tourism characteristic data; defining each cultural activity and scenic spot as a node in the graph; establishing a multi-dimensional edge connection system for the graph; based on the constructed graph structure, fusing the geographic features and the time features into a multi-scale high-dimensional spatial-temporal feature space; constructing a hierarchical reinforcement learning framework; introducing a multi-target reward function into the reinforcement learning framework, wherein the multi-target reward function comprises a culture experience integrity index, a space-time coordination index and a risk avoidance index; and receiving a culture type and a scenic spot type which the user wants to embody, and generating an optimal route and travel time which meet the requirements of the user through the hierarchical reinforcement learning framework in the spatial-temporal feature space. The method is especially suitable for tourist route planning with rich local culture and complex geographical environment in Sichuan and Tibet areas and the like.

Description

Technical Field

[0001] The present invention belongs to the field of tourism planning, and in particular relates to a tourism service method and system that integrates local culture and tourism characteristics. Background Art

[0002] With the rapid development of the global tourism industry, personalized travel services are becoming a mainstream market demand. This is especially true in unique geographical locations like the Sichuan-Tibet Highway, where travelers not only desire to enjoy the unique natural landscape but also deeply experience the local culture. However, existing itinerary recommendation methods typically focus solely on location or transportation access, rarely considering multi-dimensional factors such as cultural experience, travel safety, and time and space constraints.

[0003] Furthermore, the Sichuan-Tibet region faces complex geography, including high altitude, severe weather, and road hazards, all of which can significantly impact travel safety. Traditional travel recommendation systems often fail to respond promptly to sudden geological disasters or weather changes, limiting tourists' travel experiences.

[0004] When it comes to cultural experiences, existing systems have relatively simple recommendation strategies, primarily based on users' historical preferences, and fail to deeply explore local cultural characteristics. Because local cultural activities are often time-sensitive and spatially distributed, traditional methods struggle to provide dynamically adjusted, personalized recommendations tailored to tourists' preferences.

[0005] At the same time, current path planning methods are mostly based on static shortest path algorithms, which ignore factors such as travel time arrangements, in-depth cultural experience, and environmental risks, making it difficult to provide tourists with the best travel plans. Summary of the Invention

[0006] In order to solve the problems in the prior art, the present invention provides a tourism service method that integrates local culture and tourism characteristics, comprising the following steps:

[0007] Collecting local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities;

[0008] Collecting tourism characteristic data, wherein the tourism characteristic data at least includes characteristics, geographical location, transportation information, and risk information of the scenic spot;

[0009] Each cultural activity and attraction is defined as a node in the graph, and each node has a feature vector including time feature, cultural feature, and geographical location;

[0010] Establishing a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges;

[0011] Based on the constructed graph structure, the spatiotemporal graph convolutional network is used to simultaneously learn features in the spatial and temporal dimensions. Spatial convolution is used to extract geographical features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time. Fusion convolution is used to fuse the geographical features and temporal features into a multi-scale high-dimensional spatiotemporal feature space.

[0012] A hierarchical reinforcement learning framework is constructed. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization.

[0013] A multi-objective reward function is introduced into the reinforcement learning framework, including: cultural experience integrity index, spatiotemporal coordination index, and risk aversion index;

[0014] The user's desired cultural type and scenic spot type are received, and the optimal route and travel time that meet the user's needs are generated in the spatiotemporal feature space through the hierarchical reinforcement learning framework.

[0015] The present invention also provides a tourism service system integrating local culture and tourism characteristics, including the following modules:

[0016] A first collection module is used to collect local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities;

[0017] The second collection module is used to collect tourism characteristic data, wherein the tourism characteristic data at least includes the characteristics, geographical location, transportation information, and risk information of the scenic spot;

[0018] A node establishment module is used to define each cultural activity and scenic spot as a node in the graph, and each node has a feature vector including time feature, cultural feature, and geographical location;

[0019] An edge connection module, configured to establish a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges;

[0020] The feature fusion module is used to simultaneously learn features in spatial and temporal dimensions using a spatiotemporal graph convolutional network based on the constructed graph structure. Spatial convolution is used to extract geographic features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time. Fusion convolution is used to fuse the geographic features and temporal features into a multi-scale high-dimensional spatiotemporal feature space.

[0021] The machine learning module is used to build a hierarchical reinforcement learning framework. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization.

[0022] A reward module is used to introduce a multi-objective reward function into the reinforcement learning framework, including: a cultural experience integrity index, a spatiotemporal coordination index, and a risk aversion index;

[0023] The planning module is used to receive the cultural type and attraction type that the user wants to experience, and generate the optimal route and travel time that meets the user's needs in the spatiotemporal feature space through the hierarchical reinforcement learning framework.

[0024] This paper proposes a personalized tourism service method and system that integrates local culture and tourism characteristics. Based on multimodal data fusion, spatiotemporal feature learning, and hierarchical reinforcement learning optimization, this method can effectively address the limitations of existing tourism route planning methods. This method has the following beneficial effects:

[0025] By constructing a multi-dimensional spatiotemporal feature space based on cultural characteristics, event times, and geographic location, we ensure that tourists can fully experience local cultural activities, including religious ceremonies, folk festivals, and historical sites, within a limited timeframe. We also employ a cultural theme clustering strategy to intelligently recommend customized itineraries based on tourists' individual cultural preferences (such as Tibetan Buddhism, folk culture, or natural scenery), maximizing the depth of their cultural experience.

[0026] Through a spatiotemporal graph convolutional network, deep learning of node features across both time and space ensures that users visit attractions and activities that meet their preferences within a reasonable time window. This effectively addresses resource waste caused by factors such as time conflicts and spatial redundancy, ensures that route planning is highly coordinated in both time and space, and improves overall travel efficiency.

[0027] By incorporating geographic risk edges and risk aversion indicators, travel routes are dynamically adjusted based on real-time weather, traffic conditions, and geological disaster warnings to ensure safe travel. A hierarchical reinforcement learning framework updates the route selection strategy in real time, prioritizing travel plans in high-risk areas to avoid potential travel risks.

[0028] The present invention is particularly suitable for tourism route planning in areas with rich local culture and complex geographical environment, such as the Sichuan-Tibet region, and is especially suitable for complex travel scenarios that require personalized cultural experience and dynamic risk avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0032] Below, the invention is preferably described with reference to the accompanying drawings and specific embodiments.

[0033] This embodiment solves the above problem through the following steps:

[0034] In one embodiment, reference Figure 1 The present invention provides a tourism service method that integrates local culture and tourism characteristics, involving a personalized tourism route planning method based on multimodal data fusion, spatiotemporal feature learning and hierarchical reinforcement learning optimization. It aims to comprehensively consider the timeliness characteristics of local cultural activities, the complexity of the geographical environment of scenic spots and the personalized needs of users, and generate an optimal tourism route that meets the user's cultural preferences, tourism interests and actual travel constraints by constructing a multidimensional spatiotemporal feature map and combining it with a reinforcement learning algorithm, thereby maximizing the depth of tourists' cultural experience and the quality of tourism services while ensuring travel efficiency and safety.

[0035] Specifically, the method of this embodiment includes the following steps:

[0036] Step S10: collecting local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities.

[0037] In an embodiment of the present invention, step S10 involves collecting data related to local cultural activities. The local cultural data refers to a collection of information reflecting the unique cultural characteristics and social and historical significance of a specific region. The information can be used to describe the basic attributes, occurrence time, geographical location and cultural connotation of cultural activities, aiming to provide personalized recommendation basis for tourism services based on local cultural characteristics.

[0038] The local cultural data shall at least include the following:

[0039] Cultural activity time refers to the time window in which a cultural activity occurs, including its start time, end time, duration, and periodicity. For example, in the Sichuan-Tibet region, Tibetan New Year is a highly culturally significant festival. Its timing is adjusted according to the Tibetan lunar calendar and has a fixed periodic window, typically starting on the first day of the first lunar month and lasting 15 days. The system needs to record the specific timestamps of these festivals to facilitate time-sensitive matching in travel route planning.

[0040] The location of a cultural event refers to the actual geographic location where the event takes place, typically represented by latitude and longitude coordinates, the name of an administrative region, or the name of a specific scenic spot. For example, the renowned Shoton Festival in the Sichuan-Tibet region is primarily held in Lhasa's Drepung Monastery and Norbulingka. This location information includes not only the location of the core event area but also includes additional information such as surrounding cultural venues, participating areas, and transportation accessibility.

[0041] In addition, local cultural data may also include the following extended information (optional collection based on implementation requirements):

[0042] Cultural activity types, such as religious ceremonies, folk performances, ethnic festivals, and intangible cultural heritage displays, are used to categorize and thematically identify cultural activities. The Jokhang Temple circumambulation ceremony in the Sichuan-Tibet region is an example of a religious cultural activity.

[0043] Participation conditions include participation thresholds (e.g. whether advance reservations are required, whether it is open to the public), participant number limits (e.g. the participant number threshold for large-scale Dharma assemblies at Serthar Buddhist Institute), etc.

[0044] The cultural value level assigns a cultural value intensity coefficient to an activity based on its historical status, cultural influence, and social participation. For example, the cycling cultural activities on the Ancient Tea-Horse Road can be assessed as high cultural value activities based on their historical heritage significance and ecological impact.

[0045] By collecting the above-mentioned local cultural data, the present invention can provide a sufficient cultural feature basis for subsequent travel route planning, ensuring that travel recommendations not only meet travel convenience in space, but also are highly consistent with the actual time of cultural activities, thereby providing users with a personalized travel experience that deeply integrates local culture.

[0046] Step S20: collecting tourism characteristic data, wherein the tourism characteristic data at least includes characteristics, geographical location, traffic information, and risk information of the scenic spot.

[0047] In this embodiment of the present invention, step S20 involves collecting data related to tourism characteristics. This data refers to a multi-dimensional collection of information describing the uniqueness, geographic distribution, accessibility, and potential risks of specific tourist attractions or resources. This data provides a critical foundation for generating personalized itineraries that meet user needs, ensuring that the route planning process balances the richness of the travel experience with safety and feasibility.

[0048] The tourism characteristic data at least includes the following contents:

[0049] The characteristics of a scenic spot refer to the characteristic descriptions that can reflect the unique attraction and cultural connotations of the scenic spot, including but not limited to natural landscape characteristics, cultural and historical background, ecological environment characteristics, and entertainment functions.

[0050] For example, in the Sichuan-Tibet region, the Daocheng Yading Scenic Area is famous for its pristine natural beauty, snow-capped mountains, meadows and lakes, and is known as the "last pure land on the blue planet." The characteristic data of this attraction include landscape type (snow-capped mountains, grasslands), viewing season (best in autumn), photographic value, and ecological environment protection level.

[0051] Geographical location refers to the specific location information of a scenic spot in space, including latitude and longitude coordinates, administrative region affiliation (such as province, city, county), altitude and surrounding geographical environment characteristics.

[0052] For example, the Mount Everest Base Camp is located in Dingri County, Shigatse City, Tibet. Its geographic location data should include detailed coordinates (27.9881°N, 86.9250°E), altitude (approximately 5,200 meters), and spatial distance from other important nodes (such as Lhasa City or Zhangmu Port).

[0053] Traffic information refers to data describing the traffic conditions required for tourists to reach a certain attraction, including but not limited to accessible routes, types of transportation, travel time consumption, real-time traffic flow and road accessibility.

[0054] For example, traffic information to Nyingchi Peach Blossom Valley needs to include the main traffic routes from Lhasa (such as National Highway 318), available transportation methods (self-driving, chartered car, shuttle bus), real-time traffic conditions (traffic congestion level, road maintenance status) and potential traffic control information.

[0055] Risk information refers to potential risk data that tourists may encounter during their travel or tour, including but not limited to:

[0056] Natural environmental risks, such as the risk of altitude sickness in the Sichuan-Tibet Plateau and the probability of sudden landslides or mud-rock flows;

[0057] Weather risks, including severe weather warnings within specific time periods (such as heavy snow, heavy fog, sandstorms, etc.);

[0058] Traffic safety risks, such as landslides, accident-prone sections, bridge maintenance, and other traffic obstructions;

[0059] Tourist carrying risk refers to the problem of reduced safety and experience quality caused by the actual number of tourists at a scenic spot exceeding the reception capacity threshold.

[0060] For example, when traveling to Serthar Buddhist Academy, the risk information includes potential risks such as icy roads (more common in winter), the possibility of altitude sickness (because the attraction is about 4,000 meters above sea level), the risk of traffic congestion (during large-scale religious activities), and a shortage of accommodation resources.

[0061] By collecting the above-mentioned tourism characteristic data, the present invention can provide users with personalized route planning within a specific time and dynamically adjust travel plans according to real-time environmental information, thereby ensuring the quality of tourism experience and travel safety.

[0062] Step S30: define each cultural activity and scenic spot as a node in the graph, and each node has a feature vector including a time feature, a cultural feature, and a geographical location.

[0063] This step represents the cultural activities and attractions to be modeled as nodes in a graph structure and assigns a feature vector with multi-dimensional attributes to each node. These feature vectors include, but are not limited to, temporal features, cultural characteristics, and geographic location information, enabling relationship modeling and feature propagation based on graph neural networks or other graph-based algorithms. This approach effectively captures the potential relationships between cultural activities and attractions, promotes information flow within complex network structures, and provides a data foundation for subsequent recommendations, traffic forecasting, and route planning.

[0064] A node is the basic unit in a graph structure, representing a specific entity or object. In this step, cultural activities and attractions are defined as nodes. Each node represents a specific location or cultural event and has unique semantic information and attributes.

[0065] A feature vector is a multidimensional numerical vector that describes the properties and characteristics of a node. In this method, the feature vector of each node consists of the following parts:

[0066] Temporal characteristics: including the time of occurrence, duration, seasonality, opening hours, etc. of the activity;

[0067] Cultural characteristics: including activity type (such as festivals, exhibitions, performances), cultural background, historical significance, etc.;

[0068] Geographic location: includes spatial coordinate information such as longitude, latitude, and altitude.

[0069] A graph is a data structure consisting of nodes and edges, used to represent relationships between entities. In this approach, nodes represent cultural activities and attractions, and edges represent relationships between different nodes (such as geographical proximity, thematic similarity, or temporal association).

[0070] The specific implementation steps of step S30 include:

[0071] Step S31. Define nodes

[0072] a) Definition of cultural activity nodes

[0073] Each cultural activity is defined as an independent node and assigned a unique identifier (ID).

[0074] Assign activity attributes to each node, including activity name, type, time range, and cultural category.

[0075] b) Definition of scenic spot nodes

[0076] Each tourist attraction (e.g., museum, historical site, natural landscape) is defined as an independent node and assigned a unique identifier (ID).

[0077] Assign attraction attributes to each node, including opening hours, theme categories, visitor capacity, etc.

[0078] Step S32. Construct feature vector

[0079] a) Temporal feature coding

[0080] Periodic time encoding (e.g., sine / cosine functions) is used to convert the temporal attributes of activities into continuous numerical features, recording the duration of events and normalizing them to values ​​between 0 and 1.

[0081] b) Cultural characteristics coding

[0082] Use one-hot encoding to classify cultural types, for example:

[0083] 0 means historical activity,

[0084] 1 means artistic performance,

[0085] 2 indicates festival activities.

[0086] The cultural context of the activity is embedded and encoded, and the text description is mapped into a fixed-length vector to capture the latent semantic information.

[0087] c) Geographic location information encoding

[0088] The longitude, latitude and altitude of the scenic spot are directly input as numerical features;

[0089] Calculate the distance to other nodes and build a spatial adjacency matrix to define the spatial relationship between nodes.

[0090] In this step, cultural activities and attractions are abstracted as nodes in a graph structure, which can effectively represent the complex spatial, temporal, and cultural relationships between them. By constructing feature vectors, temporal features, cultural features, and geographic location information are integrated into a unified numerical representation, which facilitates model understanding and processing.

[0091] Step S40: establishing a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges.

[0092] Based on the constructed graph structure, this step establishes a multi-dimensional edge connection system between nodes to describe the relationship between cultural activities and attractions in multiple dimensions, such as geographic space, risk factors, cultural connections, and time-space constraints. The edge connection system includes but is not limited to geographic distance edges, geographic risk edges, cultural association edges, and time-space constraint edges. Each type of edge is measured by weight, and the weight coefficient is adjusted according to the actual application scenario. Through this step, the complex multi-dimensional relationships between nodes can be fully expressed, improving the performance of graph neural networks in feature propagation and information aggregation, thereby providing data support for three-dimensional modeling, path planning, and traffic forecasting.

[0093] An edge in a graph structure refers to the connection between two nodes and is used to describe the interaction between nodes. In this step, the edge types include:

[0094] Geographic distance edge: an edge established based on the physical distance between two nodes in spatial coordinates;

[0095] Geographic risk edge: an edge established based on geographical environment or security risks, such as natural disaster risk areas or congestion risk areas;

[0096] Culturally relevant edges: edges based on similarities in cultural themes, categories, or historical backgrounds;

[0097] Spatiotemporal constraint edge: An edge established based on temporal and spatial synchronization constraints, such as activities that occur at the same time and in close locations.

[0098] The specific implementation of step S40 may include:

[0099] Step S41. Construction of geographic distance edges

[0100] Calculate the Euclidean distance or great circle distance between two nodes (i.e. cultural activities or attractions);

[0101] If dij ≤D th , then a geographic distance edge is established between nodes i and j; the weight of the distance edge is defined as the inverse of the distance (the closer the distance, the greater the weight):

[0102]

[0103] Among them, d ij is the geographical distance between nodes i and j, D th is the preset distance value, is the geographic distance weight, and ε is a small constant to avoid division by zero.

[0104] Step S42. Construction of geographical risk edges

[0105] a) Definition of risk factors

[0106] Assign a risk score to each node based on historical data, geological environment, natural disaster information, or traffic accident data, for example:

[0107] Flood risk index, earthquake risk index; crowd density and traffic congestion.

[0108] b) Edge definition and weight assignment

[0109] If there is a shared risk factor between two nodes, a geographical risk edge is established; the weight of the risk edge is defined according to the weighted average of the node risk scores.

[0110] Step S43. Construction of cultural association edges

[0111] The cultural feature vectors of the nodes are similarly calculated (such as cosine similarity); if the similarity is greater than a preset threshold, a cultural association edge is established between the nodes.

[0112] Step S44. Construction of spatiotemporal constraint edges

[0113] a) Measure the time difference between the time feature vectors of two nodes:

[0114] Δt ij =|t i -t j |

[0115] Among them, t i and t j is the time when the activities of the two nodes occurred.

[0116] b) Spatial constraint calculation

[0117] Calculate the spatial proximity of two nodes. If the geographical distance d ij ≤D th And Δt ij≤T th (time threshold), then a spatiotemporal constraint edge is established.

[0118] c) Edge definition and weight assignment

[0119] Assign edge weights based on temporal and spatial similarity:

[0120]

[0121] Among them, σ d and σ t are the normalization parameters of distance and time.

[0122] Step S45. Construct adjacency matrix and weight matrix

[0123] Construct a multidimensional adjacency matrix A, where each dimension represents an edge type, for example:

[0124] A={A geo ,A risk ,A culture ,A st}

[0125] Assign a weight matrix W to each dimension of the adjacency matrix:

[0126] W={w geo ,w risk ,w culture ,w st}

[0127] In this step, a multi-dimensional edge connection system accurately describes the relationships between nodes across multiple dimensions, including space, risk, culture, and time, avoiding the limitations of traditional single-dimensional edge definitions. Geographic distance edges and spatiotemporal constraint edges ensure the consistency of the 3D model across both spatial and temporal dimensions, contributing to the generation of a realistic and accurate 3D model.

[0128] Step S50: Based on the constructed graph structure, a spatiotemporal graph convolutional network is used to simultaneously perform feature learning in the spatial and temporal dimensions; spatial convolution is used to extract geographic features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time; fusion convolution is used to fuse the geographic features and the temporal features into a multi-scale high-dimensional spatiotemporal feature space.

[0129] This step jointly learns the spatial and temporal features of nodes using a graph-based spatiotemporal graph convolutional network. Spatial convolution is used to extract geographic features between nodes in the graph, while temporal convolution is used to learn the temporal characteristics of node features as they evolve over time. Fusion convolution integrates these spatial and temporal features into a multi-scale, high-dimensional spatiotemporal feature space. This step aims to capture the dynamic relationships between cultural activities and attractions across different temporal and spatial dimensions, providing a spatially and temporally consistent feature representation for applications such as 3D reconstruction, traffic flow prediction, and path planning.

[0130] In this step, the spatiotemporal graph convolutional network (GCNN) is a deep learning network based on graph-structured data that can simultaneously learn features in both spatial and temporal dimensions. It combines the spatial modeling capabilities of graph convolutional networks (GNNs) with the time series processing capabilities of one-dimensional convolutional networks (1D-CNNs) or recurrent neural networks (RNNs), making it suitable for modeling complex systems with spatiotemporal dependencies.

[0131] In this step, spatial convolution refers to performing convolution operations on the spatial structure of the graph, capturing the spatial feature propagation relationship between adjacent nodes through the adjacency matrix and the node feature matrix, and mainly reflecting the interaction between nodes in the geographic space.

[0132] Temporal convolution refers to performing a one-dimensional convolution operation on node features in the time dimension to learn the dynamic evolution of node features over time, thereby capturing the trend of activities or attractions evolving over time.

[0133] Fused convolution refers to the fusion of features obtained by spatial convolution and temporal convolution, extracting high-dimensional features at different time and spatial scales through multi-scale learning (such as different convolution kernel sizes), and generating a unified spatiotemporal feature vector.

[0134] Multi-scale high-dimensional spatiotemporal feature space refers to the feature space formed by integrating feature vectors at multiple time and space scales, which can characterize the complex spatiotemporal dynamic relationships between nodes at different levels.

[0135] The specific implementation of step S50 may include:

[0136] Step S51. Constructing a spatiotemporal graph convolutional network

[0137] a) Input feature definition

[0138] The input consists of the node feature matrix and the adjacency matrix Composition, of which:

[0139] N is the number of nodes (including cultural activities and attractions);

[0140] F is the characteristic dimension of each node (including time characteristics, cultural characteristics and geographical location characteristics).

[0141] b) Time dimension modeling

[0142] Organize the node feature matrix in time series to form a three-dimensional input tensor in:

[0143] T represents the number of time steps (e.g., day, week, month);

[0144] X t Represents the node feature matrix at time step t.

[0145] Step S52: Spatial convolution implementation

[0146] a) Adjacency matrix normalization

[0147] Perform symmetric normalization on the adjacency matrix A and obtain

[0148]

[0149] Where D is the degree matrix of the node.

[0150] b) Convolution calculation formula

[0151] The spatial convolution operation formula is:

[0152]

[0153] in:

[0154] is the feature matrix output by the l-th layer spatial convolution;

[0155] is the weight matrix of the l-th layer of spatial convolution;

[0156] σ is the activation function (such as ReLU).

[0157] c) Spatial feature extraction

[0158] Extract geographical features related to node neighbors, including geographical distance, cultural similarity, and spatial risk information of adjacent nodes.

[0159] Step S53. Temporal convolution implementation

[0160] a) One-dimensional convolution operation

[0161] Perform 1D convolution operation on each node feature in the time dimension, the formula is:

[0162]

[0163] in:

[0164] Output feature matrix for the lth layer of temporal convolution;

[0165] is the weight matrix of temporal convolution.

[0166] b) Temporal feature extraction

[0167] Capture the changing trend of node features over time and extract time series features such as activity occurrence time, duration and temporal regularity.

[0168] Step S54. Fusion convolution implementation

[0169] a) Feature fusion strategy

[0170] Concatenate the features output by spatial convolution and temporal convolution:

[0171]

[0172] Among them, L is the number of network layers, H st is the fused spatiotemporal feature matrix.

[0173] b) Multi-scale learning

[0174] Extract multi-scale spatiotemporal features through convolution kernels of different sizes:

[0175] Small-size convolution kernel: used to capture short-term change trends in local time and space;

[0176] Large-size convolution kernels: used to capture long-term trends in global time and space.

[0177] c) High-dimensional feature mapping

[0178] The fusion features are mapped to the high-dimensional feature space through the fully connected layer. The formula is:

[0179] Z=σ(H st W f +b)

[0180] in:

[0181] W f is the weight matrix of the fully connected layer;

[0182] b is the bias term.

[0183] In this step, spatial convolution is used to learn the geographic features between cultural activity and scenic spot nodes, effectively modeling the spatial interactions between nodes based on geographic distance, spatial risk, and proximity. Temporal convolution is used to explore the temporal changes in node features, capturing the cyclical changes, sudden trends, and historical dependencies of cultural activities, providing effective features for dynamic prediction.

[0184] For example:

[0185] The node list includes:

[0186] Node Name Node Type Time characteristics Cultural characteristics Geographical location Chengdu Wuhou Temple Attractions Open year-round, 9:00-17:00 Three Kingdoms History and Culture (104.05,30.65,540m) Kangding Love Song Festival Activity Mid-August every year Tibetan folk customs (101.96,30.05,2560m) Basongcuo Attractions Open year-round, 9:00-18:00 Natural scenery (93.85,29.75,3480m) Jokhang Temple in Lhasa Attractions Open year-round, 8:00-18:00 Tibetan Buddhist culture (91.12,29.66,3650m) Tibetan New Year Celebrations Activity The first day of the first lunar month every year Tibetan traditional festivals (91.13,29.66,3650m)

[0187] Establish a multi-dimensional edge connection system between nodes, including:

[0188] Geographic distance edge

[0189] Connect two attractions based on their actual road distance, for example:

[0190] (230 kilometers, edge weight = 1 / 230)

[0191] Geographical risk margin

[0192] The edge weights are established by risk scores such as weather warnings, traffic congestion, and landslides. For example:

[0193] (Edge weights are increased due to higher altitude and greater risk of climate change)

[0194] Cultural Relations

[0195] Create edges between attractions or activities with similar cultural backgrounds, such as:

[0196] (Cultural similarity > 0.8, establishing a strong connection)

[0197] Space-time constraint edge

[0198] Create edges between nodes that are close in time and space, for example:

[0199] (Time overlap, spatial distance <1km, maximum weight)

[0200] Spatiotemporal Graph Convolutional Network Modeling

[0201] Spatial convolution (extracting geographic features)

[0202] Perform spatial convolution based on the geographic adjacency matrix to learn spatial location-based feature propagation between nodes, for example:

[0203] When planning the route from Chengdu to Lhasa, the geographical distance between each node and surrounding attractions and the road conditions are taken into consideration to optimize the travel route.

[0204] Temporal convolution (learning temporal dynamic features)

[0205] Perform convolutional learning on the time-varying features of activities and attractions, such as:

[0206] Predict the crowd density at the Jokhang Temple in Lhasa during the Tibetan New Year celebrations and plan ahead to avoid peak hours.

[0207] Fused convolution (spatial and temporal feature fusion)

[0208] The outputs of spatial and temporal convolution are fused to generate high-dimensional spatiotemporal features, for example:

[0209] By integrating time and space characteristics, we provide the best travel route recommendations. For example, during the Kangding Love Song Festival, we recommend tourists to visit the Kangding Love Song Memorial Hall first and then go to the surrounding natural attractions to avoid traffic jams during the peak festival period.

[0210] Step S60: construct a hierarchical reinforcement learning framework. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization.

[0211] This step constructs a hierarchical reinforcement learning framework, including an upper-level policy network, a middle-level planner, and a lower-level executor.

[0212] The upper-level strategy network is based on the Monte Carlo tree search algorithm to generate cultural theme clustering schemes for cultural activities and scenic spots nodes to guide the path planning and strategy execution of the lower level.

[0213] The mid-level planner constructs a set of feasible paths that satisfy spatiotemporal constraints through a generative adversarial network.

[0214] The lower-level executor performs fine-tuning optimization of path planning based on a double-delayed deep deterministic policy gradient algorithm to maximize the effectiveness and stability of the overall strategy in complex environments.

[0215] In this paper, the hierarchical reinforcement learning framework refers to a framework that decomposes complex decision-making tasks into multiple layers (upper, middle, and lower layers) for optimization learning. Different layers focus on decision-making tasks of different granularity, from high-level policy decisions to low-level action execution, forming a hierarchical optimization structure from global to local to detailed.

[0216] Monte Carlo Tree Search (MCTS) is a decision-making algorithm based on randomized simulation. It constructs a search tree, combines random sampling with heuristic search, and finds the optimal decision path under a specific strategy. In this method, it is used to explore the optimal clustering scheme for cultural activities and tourist attractions.

[0217] The cultural theme clustering scheme refers to clustering and classifying scenic spots and cultural activities along the Sichuan-Tibet Highway according to cultural characteristics (such as religion, folk customs, and historical background) to generate theme routes that match tourists' preferences, such as the Tibetan Buddhist cultural route and the Western Sichuan folk experience route.

[0218] A generative adversarial network (GAN) is a deep learning framework consisting of a generator and a discriminator. The generator is responsible for generating new data samples under specific conditions, while the discriminator is responsible for determining whether the samples are realistic. In this method, GAN is used to generate a set of feasible travel routes under spatiotemporal constraints.

[0219] Spatiotemporal constraints refer to the temporal and spatial restrictions of nodes, including factors such as the opening hours of scenic spots, the time of events, geographical proximity, and weather risks.

[0220] The double-delayed deep deterministic policy gradient (TD3) reinforcement learning algorithm, combined with policy delayed update and target network perturbation technology, can effectively alleviate the over-estimation problem during policy training and improve the stability and convergence speed of policy execution.

[0221] Furthermore, the specific implementation steps of step S60 include:

[0222] Step S61. Upper-layer strategy network: cultural theme clustering scheme generation (based on MCTS)

[0223] a) Build a search tree

[0224] Each node represents a cultural activity or attraction, and the edges represent the relationships between them (e.g., geographical proximity, cultural connection).

[0225] The root node represents the starting point of the trip (e.g. Chengdu), and the end node represents the end point (e.g. Lhasa).

[0226] b) Tree search process

[0227] Selection: Select child nodes from the root node based on the upper confidence limit strategy.

[0228] Extension: Expand unvisited nodes to form a new subtree.

[0229] Simulation: Perform random sampling simulation on each leaf node to evaluate the effectiveness of the clustering path.

[0230] Feedback: Update the simulation results back to the parent node to optimize future selection probabilities.

[0231] c) Clustering strategy output

[0232] The final output is a cultural theme clustering solution that meets user preferences (such as religious culture, folk culture), for example:

[0233] Route A: Chengdu → Kangding → Basongcuo → Lhasa (focused on Tibetan Buddhism)

[0234] Route B: Chengdu → Litang → Ranwu Lake → Lhasa (focused on the natural scenery of Sichuan and Tibet)

[0235] Step S62. Mid-level planner: generating feasible paths under spatiotemporal constraints (based on GAN)

[0236] a) Generator construction

[0237] Input: cultural theme clustering scheme, starting point, end point, time and space constraints (opening hours, risk level, geographical distance).

[0238] Output: A set of preliminary feasible paths that meet the time and space constraints.

[0239] b) Discriminator construction

[0240] Input: The path plan output by the generator is compared with the real historical travel path.

[0241] Output: binary classification, 1 means feasible, 0 means infeasible.

[0242] c) Adversarial training process

[0243] By minimizing the generator loss and maximizing the discriminator loss for adversarial training, a set of high-quality paths that meet spatiotemporal constraints is finally generated.

[0244] Step S63. Lower-layer executor: path fine-tuning optimization (based on TD3)

[0245] a) Reward function definition

[0246] Define the reward function based on factors such as travel time, budget constraints, risk probability, etc.:

[0247] R = w1·(1-risk score)+w2·(1-total time)+w3·cultural preference matching

[0248] Among them, w1, w2, and w3 are different weight coefficients.

[0249] b) Policy Update

[0250] Through TD3's dual-delay policy update and target network perturbation mechanism, over-estimation during policy training is avoided, thereby improving the stability of path selection.

[0251] c) Strategy fine-tuning

[0252] Make local adjustments to the set of feasible paths, for example: reduce the length of stay in high-risk areas; avoid entering popular attractions during peak hours.

[0253] In this step, a hierarchical architecture decomposes the complex path planning task into multiple layers, from global topic clustering to local path fine-tuning. This effectively reduces computational complexity and improves decision-making efficiency. This hierarchical reinforcement learning framework can adjust decision-making strategies in real time under large-scale dynamic environments, adapting to complex climate changes, traffic conditions, and emergencies on the Sichuan-Tibet Highway.

[0254] For example, follow the examples in the above steps.

[0255] Upper policy network: Generate cultural theme clustering scheme (based on MCTS)

[0256] Input conditions:

[0257] Departure Place: Chengdu

[0258] End point: Lhasa

[0259] Tourist cultural preference: Religious culture (religious activities and attractions are preferred)

[0260] Total duration: 10 days

[0261] Output results (topic clustering scheme):

[0262] Theme A (Religious and Cultural Route): Chengdu → Litang Temples → Lhasa Jokhang Temple → Tibetan New Year Celebration

[0263] Theme B (Folk Culture Route): Chengdu → Kangding Love Song Festival → Litang → Basongcuo → Lhasa

[0264] MCTS will search for the optimal topic clustering path in the search tree based on tourists’ cultural preferences and travel time constraints, giving priority to the solution that meets cultural preferences, has the shortest travel time and the optimal geographical distance.

[0265] Mid-level planner: Generates a set of feasible paths that satisfy spatiotemporal constraints (based on GAN)

[0266] Input conditions:

[0267] Theme Route: Chengdu → Litang Temples → Jokhang Temple in Lhasa → Tibetan New Year Celebration

[0268] Time and space constraints: activity time windows, scenic spot opening hours, weather warning information

[0269] Output results (feasible path set):

[0270] Route 1: Chengdu → Ya'an → Litang Temple Complex → Basongcuo → Jokhang Temple in Lhasa

[0271] Route 2: Chengdu → Kangding → Litang Temple Complex → Ranwu Lake → Jokhang Temple in Lhasa

[0272] Route 3: Chengdu → Litang Temples → Namtso Lake → Lhasa Jokhang Temple → Tibetan New Year Celebration

[0273] The generator generates feasible paths based on the input conditions, and the discriminator screens them by comparing historical travel data and spatiotemporal constraints to ensure that each path is feasible in the actual environment.

[0274] Lower-level executor: path fine-tuning optimization (based on TD3)

[0275] Input conditions:

[0276] Set of feasible paths (output from the mid-level planner)

[0277] Objective function: minimize travel time, avoid high risks, and maximize cultural preferences

[0278] Output results (path after fine-tuning optimization):

[0279] Optimized route: Chengdu → Ya'an → Litang Temple Complex → Ranwu Lake → Jokhang Temple in Lhasa → Tibetan New Year Celebration

[0280] Estimated total duration: 9 days

[0281] Total budget: RMB 9,000

[0282] TD3 dynamically adjusts routes based on travel time, risk factors, and cultural preferences, such as shortening stays in high-risk areas or increasing time spent at religious sites while avoiding peak periods.

[0283] Step S70: introducing a multi-objective reward function into the reinforcement learning framework, including: a cultural experience integrity index, a spatiotemporal coordination index, and a risk aversion index.

[0284] This step constructs a multi-objective reward function within the reinforcement learning framework to guide the hierarchical reinforcement learning model in making path decisions under different optimization objectives. This objective reward function consists of the following three core indicators:

[0285] The cultural experience completeness index is used to quantify the completeness of a travel route in terms of cultural activities and attractions, and to measure whether the trip includes cultural elements preferred by users, such as Tibetan Buddhist culture and western Sichuan folk activities.

[0286] The spatiotemporal coordination index measures the degree of coordination between the opening hours of attractions, the time of events and geographical proximity, avoiding resource waste due to time conflicts or spatial duplications and optimizing overall travel efficiency.

[0287] The risk avoidance indicator measures the effectiveness of travel routes in avoiding potential geographical risks (such as landslides, extreme weather, road interruptions, etc.), ensuring travel safety and reducing uncertainty.

[0288] Furthermore, the implementation of step S70 includes:

[0289] Step S71. Determine the multi-objective reward function

[0290] The three core indicators are combined into the following multi-objective reward function:

[0291] R total =w1·R cultural +w2·R spatio +w3·R risk

[0292] in:

[0293] w1, w2, and w3 are weight coefficients of different indicators, reflecting the relative importance of each goal in travel decision-making;

[0294] R cultural rewards for completeness of cultural experience;

[0295] R spatio Reward for space-time coordination;

[0296] R risk Rewards for risk aversion.

[0297] Step S72: Determine the cultural experience integrity index

[0298]

[0299] in:

[0300] N visited Indicates the number of activities or attractions visited during the trip that match the user's cultural preferences;

[0301] N total Indicates the total number of activities or attractions that meet your preferences;

[0302] S similarity Scoring cultural theme similarity (e.g. using cosine similarity metric);

[0303] λ is the weighting coefficient of the similarity score.

[0304] Nodes (attractions and activities) are feature-encoded using a graph neural network. Cultural similarity between users' cultural preferences and visited nodes is calculated. During reinforcement learning training, paths that cover more cultural activities and attractions are prioritized.

[0305] The cultural experience integrity index ensures that users have a complete and diverse cultural experience during their travels, meeting their needs for cultural depth and thematic integrity.

[0306] Step S73: Determine the spatiotemporal coordination index

[0307]

[0308] in:

[0309] T confilict Indicates the total duration of activities that have time conflicts;

[0310] T total Indicates the total journey time;

[0311] D redundant Indicates the redundant path length in space;

[0312] D total Indicates the total distance of the entire trip.

[0313] During implementation, a time window is established based on the event time and the opening hours of the scenic spots. The geographical distance between the scenic spots is calculated and an adjacency matrix is ​​established. During the path generation process, routes that are coordinated in time and space are given priority.

[0314] The spatiotemporal coordination index can effectively avoid time conflicts and waste of spatial resources, ensure efficient and coherent travel plans, and save time and budget costs.

[0315] Step S74: Determine the risk aversion index

[0316]

[0317] in:

[0318] R avg represents the average risk score of the selected paths;

[0319] R max Indicates the highest risk score value in the route.

[0320] During implementation, risk scores are dynamically updated based on meteorological data, historical traffic records, and geographic risk distribution maps. During route generation, routes are dynamically adjusted based on real-time risks (e.g., avoiding landslide areas or areas with inclement weather). During reinforcement learning training, the probability of selecting high-risk nodes is reduced.

[0321] Risk aversion indicators can ensure safety during travel, reduce the interference of emergencies on travel plans, and provide reliable and stable route planning solutions.

[0322] Step S75: Determine the dynamic weight adjustment strategy

[0323] The weights w1, w2, and w3 of the three reward indicators are dynamically adjusted according to traveler preferences and actual environmental conditions.

[0324] If the user prefers cultural experience → increase w1;

[0325] If time is limited or budget is tight → increase w2;

[0326] If there is high risk weather or special environmental conditions → increase w3.

[0327] Step S70 introduces a multi-objective reward function within the reinforcement learning framework, comprehensively considering multiple optimization objectives, including cultural experience integrity, spatiotemporal coordination, and risk aversion, thereby effectively balancing travelers' cultural needs, travel efficiency, and safety. This approach not only adapts to personalized user needs but also improves the model's adaptability in complex environments. It is particularly suitable for regions rich in local culture and tourism, such as Sichuan-Tibet.

[0328] Step S80: receiving the cultural type and scenic spot type that the user wants to experience, and generating the optimal route and travel time that meets the user's needs through the hierarchical reinforcement learning framework in the spatiotemporal feature space.

[0329] This step automatically generates the optimal route and travel time arrangement that meets the user's needs by accepting the cultural type and attraction type input by the user as constraints and combining the previously constructed spatiotemporal feature space and hierarchical reinforcement learning framework.

[0330] During this process, the hierarchical reinforcement learning framework is optimized based on a multi-objective reward function. Through the decision-making mechanism of the three-layer structure of the upper-level policy network, the middle-level planner and the lower-level executor, the route is globally planned, the path is constructed under spatiotemporal constraints, and the details are fine-tuned, thereby ensuring that the route conforms to user preferences, meets spatiotemporal coordination requirements, and avoids potential risks.

[0331] Furthermore, step S80 may specifically include:

[0332] Step S81: User input reception

[0333] Input conditions include:

[0334] Cultural preferences: Religious culture (Tibetan Buddhism), folk culture (Kangding Love Song Festival), historical culture (Three Kingdoms culture), natural scenery (plateau lakes, snow-capped mountains)

[0335] Attraction type preference: such as museums, temples, natural scenic spots, festivals, etc.

[0336] Travel restrictions: total duration, budget constraints, transportation preferences, etc.

[0337] Sample input:

[0338] User preference: religious culture + natural landscape

[0339] Attraction type: Temple + Lake

[0340] Total travel days: 8 days

[0341] Budget limit: 8,000 RMB

[0342] Step S82: Upper-layer strategy network (generating topic clusters based on Monte Carlo tree search)

[0343] a) Node screening

[0344] According to the culture type and attraction type input by the user, the nodes (i.e., attractions or activities) that meet the conditions are filtered from the spatiotemporal feature space.

[0345] b) Cultural theme clustering

[0346] Cultural clustering based on similarity of user preferences, for example:

[0347] Route A (Religious and Cultural Route): Chengdu → Litang Temples → Jokhang Temple in Lhasa → Tibetan New Year Celebration

[0348] Route B (Natural Scenery Route): Chengdu → Kangding Mugecuo → Ranwu Lake → Namtso

[0349] c) Topic Prioritization

[0350] Monte Carlo tree search (MCTS) was used to simulate different routes, evaluate the cultural integrity score of each route, and select the optimal topic clustering scheme based on the simulation results.

[0351] Step S83. Mid-level planner (generating feasible paths based on generative adversarial networks)

[0352] a) Constructing constraints

[0353] Set spatiotemporal constraints based on attraction opening hours, event occurrence times, traffic conditions, and user time budget.

[0354] b) Path generation

[0355] Use the generator to generate a set of candidate paths that meet the spatiotemporal constraints.

[0356] The discriminator is used to filter out feasible paths and ensure that each path is executable in the real environment.

[0357] c) Output of feasible path set

[0358] Generate multiple feasible paths that meet the user's cultural type and attraction type requirements.

[0359] Step S84. Lower-layer executor (path fine-tuning optimization based on TD3)

[0360] a) Define the reward function

[0361] Reward function comprehensive considerations:

[0362] Cultural experience integrity (coverage of the completeness of user-specified cultural types)

[0363] Time and space coordination (rational time arrangement, avoiding conflicts and redundancies)

[0364] Risk avoidance (avoiding high-risk areas and ensuring travel safety)

[0365] b) Path detail adjustment

[0366] Dynamically adjust the node visiting order to optimize route length and travel time.

[0367] Re-route based on real-time risks (such as weather changes).

[0368] c) Output the optimal route and travel time

[0369] Provide the final travel route and detailed schedule that meets user needs.

[0370] For example:

[0371] User input:

[0372] Cultural type: Tibetan Buddhism + natural scenery

[0373] Attraction type: Temple + Plateau Lake

[0374] Travel restrictions: Total itinerary 8 days, budget 9000 RMB

[0375] Output (route planning):

[0376] Chengdu → Kangding Mugecuo → Litang Temple Group → Ranwu Lake → Lhasa Jokhang Temple → Namtso

[0377] Dynamic adjustments are made to avoid traffic risks caused by weather changes along the Kangding to Litang line.

[0378] Travel schedule:

[0379] Daily route recommendations (including departure time, arrival time, and estimated stay time) are dynamically adjusted based on user feedback and real-time weather conditions.

[0380] Step S80 utilizes a hierarchical reinforcement learning framework to generate the optimal route and travel time by combining the user-input cultural and attraction types with spatiotemporal features. This method not only customizes travel itineraries based on user preferences but also dynamically optimizes routes based on spatiotemporal constraints and risk predictions, ensuring the integrity of the cultural experience, spatial and temporal coordination, and travel safety. This approach is particularly suitable for travel in regions like Sichuan and Tibet, which are rich in local culture and tourism characteristics but also prone to numerous risk factors.

[0381] See also Figure 2 In another embodiment, the present invention further provides a tourism service system integrating local culture and tourism characteristics, comprising:

[0382] A first collection module is used to collect local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities;

[0383] The second collection module is used to collect tourism characteristic data, wherein the tourism characteristic data at least includes the characteristics, geographical location, transportation information, and risk information of the scenic spot;

[0384] A node establishment module is used to define each cultural activity and scenic spot as a node in the graph, and each node has a feature vector including time feature, cultural feature, and geographical location;

[0385] An edge connection module, configured to establish a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges;

[0386] The feature fusion module is used to simultaneously learn features in spatial and temporal dimensions using a spatiotemporal graph convolutional network based on the constructed graph structure. Spatial convolution is used to extract geographic features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time. Fusion convolution is used to fuse the geographic features and temporal features into a multi-scale high-dimensional spatiotemporal feature space.

[0387] The machine learning module is used to build a hierarchical reinforcement learning framework. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization.

[0388] A reward module is used to introduce a multi-objective reward function into the reinforcement learning framework, including: a cultural experience integrity index, a spatiotemporal coordination index, and a risk aversion index;

[0389] The planning module is used to receive the cultural type and attraction type that the user wants to experience, and generate the optimal route and travel time that meets the user's needs in the spatiotemporal feature space through the hierarchical reinforcement learning framework.

[0390] It should be noted that the explanation of the above-mentioned embodiment of the tourism service method that integrates local culture and tourism characteristics is also applicable to the device of the embodiment of the present application and will not be repeated here.

[0391] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0392] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0393] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0394] The above is only a specific implementation method of the present application. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be included in the protection scope of this application. The protection scope of this application should be based on the protection scope of the claims. For some module structures that are not particularly clear in the present invention, the content recorded in the prior art shall prevail. The prior art mentioned in the above background technology section and the specific embodiment section of the present invention can be regarded as part of the present invention and used to understand the meaning of some technical features or parameters.

Claims

1. A tourism service method integrating local culture and tourism characteristics, characterized in that: The method comprises the following steps: Collecting local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities; Collecting tourism characteristic data, wherein the tourism characteristic data at least includes characteristics, geographical location, transportation information, and risk information of the scenic spot; Each cultural activity and attraction is defined as a node in the graph, and each node has a feature vector including time feature, cultural feature, and geographical location; Establishing a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges; Based on the constructed graph structure, the spatiotemporal graph convolutional network is used to simultaneously learn features in the spatial and temporal dimensions. Spatial convolution is used to extract geographical features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time. Fusion convolution is used to fuse the geographical features and temporal features into a multi-scale high-dimensional spatiotemporal feature space. A hierarchical reinforcement learning framework is constructed. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization. A multi-objective reward function is introduced into the reinforcement learning framework, including: cultural experience integrity index, spatiotemporal coordination index, and risk aversion index; The user's desired cultural type and scenic spot type are received, and the optimal route and travel time that meet the user's needs are generated in the spatiotemporal feature space through the hierarchical reinforcement learning framework.

2. The tourism service method integrating local culture and tourism characteristics according to claim 1 is characterized in that: The spatial convolution refers to performing a convolution operation on the spatial structure of a graph, capturing the spatial feature propagation relationship between adjacent nodes through the adjacency matrix and the node feature matrix.

3. The tourism service method integrating local culture and tourism characteristics according to claim 1, characterized in that: The temporal convolution refers to performing a one-dimensional convolution operation on node features in the time dimension to learn the dynamic evolution law of node features over time.

4. The tourism service method integrating local culture and tourism characteristics according to claim 1, characterized in that: The fused convolution refers to fusing the features obtained by spatial convolution and temporal convolution, extracting high-dimensional features at different time and spatial scales through different convolution kernel sizes, and generating a unified spatiotemporal feature vector.

5. The tourism service method integrating local culture and tourism characteristics according to claim 1, characterized in that: The cultural theme clustering scheme refers to clustering and classifying attractions and cultural activities along the route according to cultural characteristics to generate theme routes that match tourists' preferences.

6. A tourism service system integrating local culture and tourism characteristics, characterized by: The system includes the following modules: A first collection module is used to collect local cultural data, wherein the local cultural data at least includes the time and location of local cultural activities; The second collection module is used to collect tourism characteristic data, wherein the tourism characteristic data at least includes the characteristics, geographical location, transportation information, and risk information of the scenic spot; A node establishment module is used to define each cultural activity and scenic spot as a node in the graph, and each node has a feature vector including time feature, cultural feature, and geographical location; An edge connection module, configured to establish a multi-dimensional edge connection system for the graph, including: geographical distance edges, geographical risk edges, cultural association edges, and spatiotemporal constraint edges; The feature fusion module is used to simultaneously learn features in spatial and temporal dimensions using a spatiotemporal graph convolutional network based on the constructed graph structure. Spatial convolution is used to extract geographic features between nodes, and temporal convolution is used to learn the temporal features of node features evolving over time. Fusion convolution is used to fuse the geographic features and temporal features into a multi-scale high-dimensional spatiotemporal feature space. The machine learning module is used to build a hierarchical reinforcement learning framework. The upper-level policy network generates a cultural theme clustering scheme based on Monte Carlo tree search. The middle-level planner uses a generative adversarial network to construct a set of feasible paths under spatiotemporal constraints. The lower-level executor uses a double-delayed deep deterministic policy gradient algorithm for fine-tuning and optimization. A reward module is used to introduce a multi-objective reward function into the reinforcement learning framework, including: a cultural experience integrity index, a spatiotemporal coordination index, and a risk aversion index; The planning module is used to receive the cultural type and attraction type that the user wants to experience, and generate the optimal route and travel time that meets the user's needs in the spatiotemporal feature space through the hierarchical reinforcement learning framework.

7. The tourism service system integrating local culture and tourism characteristics according to claim 6 is characterized in that: The spatial convolution refers to performing a convolution operation on the spatial structure of a graph, capturing the spatial feature propagation relationship between adjacent nodes through the adjacency matrix and the node feature matrix.

8. The tourism service system integrating local culture and tourism characteristics according to claim 6 is characterized in that: The temporal convolution refers to performing a one-dimensional convolution operation on node features in the time dimension to learn the dynamic evolution law of node features over time.

9. The tourism service system integrating local culture and tourism characteristics according to claim 6 is characterized in that: The fused convolution refers to fusing the features obtained by spatial convolution and temporal convolution, extracting high-dimensional features at different time and spatial scales through different convolution kernel sizes, and generating a unified spatiotemporal feature vector.

10. The tourism service system integrating local culture and tourism characteristics according to claim 6, characterized in that: The cultural theme clustering scheme refers to clustering and classifying attractions and cultural activities along the route according to cultural characteristics to generate theme routes that match tourists' preferences.

Images